Determination of Optimal Measurement Guardbands

Abstract

Depending on the capability of the measurement process, guardbands are often applied to key characteristics of the final product. Measurement guardbands can be set inside the specification limits to better assure that only items within the guardbands are shipped to the customer. In general, optimal measurement guardbands are determined by taking the risks of incorrect decisions into account. However, the measurements generated from inspecting such product characteristics do not necessarily conform to the usual normality assumption. Since the product characteristic is a random variable that is measured with non-negligible error, the measurements do not estimate the true product characteristic density, but rather its convolution with the measurement error density. In this paper, we consider a kernel estimator for the density of the measurements and a deconvolution estimator of the product characteristic density. The estimators can be used in conjunction with the risks to determine the optimal guardbands.